LTV:CAC deep dive part 2: Measurement

Introduction

LTV:CAC (lifetime value to customer acquisition cost ratio) is a very popular marketing efficiency metric and captures the unit economics of every business. Despite its popularity though, it can get confusing when it comes to understanding its variations, calculations and interpretations. So in this multi-part post I discuss from first principles the following topics:

• Part 1: Misconceptions (covered here)

• Part 2: Measurement framework

• Part 3: Optimal LTV:CAC ratio

• Part 4: LTV:CAC & marketing spend optimisation

Welcome to the second part of the series!

The role of LTV

Let’s be honest, the protagonist of this metric has to be CAC. Partly because LTV:CAC is a marketing metric and while LTV will be a function of your product, CAC is pure marketing. In addition, many LTV formulas include CAC, so in this case you can’t have LTV without computing CAC first. Nonetheless, LTV is a major part of this metric and we should discuss its role which will determine its definition in the context of marketing. It is a standard practice to use different variants of LTV based on the needs of the business.

LTV enters the calculation as a benchmark that makes CAC comparable across various marketing campaigns and, when calculated on aggregate, across multiple organisations. A CAC of $100 might be detrimental for a food delivery app, but it can be considered a “deal” for Apple. When comparing marketing efficiency across channels like PR or social media content to paid search, you also have to combine CAC with LTV estimates that would be relevant to the customers coming from these channels.

Looking at CAC in isolation doesn’t tell us the full story.

Properties of LTV

A good measure of efficiency

Let’s consider a marketplace like Amazon. Marketplaces might sell extremely expensive items but they don’t get to keep all the money (GMV vs Revenue). So buying an expensive TV might cost $1000 but only a small fraction will go to the marketplace. Let’s assume for this example that the marketplace would generate $10 out of this sale. So a CAC of $100, while only 10% of GMV, would be disastrous for the marketplace business that only gets to keep $10. So you have to establish prioritisation when it comes to who gets money and when. If selling a product is associated with some COGS that are necessary for the product to reach your customer, you need to take them into account in your LTV. For instance, if you have an iOS app, Apple will get 30% of your revenue at checkout. Basically, any variable costs that are required for servicing the customer have to be included in LTV. This is also know as contribution margin per customer as it is the way the customer contributes to covering fixed costs and generating profit for the business.

So in a cohort of n customers, LTV for a time window of T would be:

Accounting for uncertainty

Money today is always more valuable than money a year from now (time value of money) due to its earning potential. For the above reasons, we should be applying a discount rate in the formula. Each business will have a required rate of return that is determined by their capacity to generated value from various ways of deploying capital. For a business with high required rate of return, and thus many options for value generation, only efficient marketing opportunities would be justified.

LTV:CAC variations

Short vs long run

I would like to take a step back and reason over the necessity for these variations. The value of a metric should be prescriptive. A prescription in this case can be useful to support decision making in:

• the short run (Channel LTV:CAC, CAC includes advertising costs)

• the long run (All-In LTV:CAC, CAC in addition includes salaries, tools, measurement costs)

The reason we distinguish between the short and long run is because variables that are considered fixed in the short run, become variable in the long run (e.g. salaries) and depending on the marketer’s place in the business hierarchy, they will be considering different levers, e.g. lower-in-the-hierarchy marketers would optimise for marketing spend given budget constraint (short run decision making), while a CMO would optimise for salaries among other things (long run decision making).

Observe how in Example 2 from part 1 we used monthly data in order to estimate CAC based on advertising variable costs only. This CAC would fall under the Channel CAC category. Usually, looking at marketing data on monthly or weekly granularity would be a reasonable thing to do, given that the marketing team is experimenting with various channels to create the necessary data for the estimation process to take place. This is the price you have to pay if you want to accurately know your CAC.

For the All-In CAC the only change would be the addition of fixed (in the short term) costs in the calculation of total costs in the numerator.

Reporting vs Optimisation

Another important consideration is whether you are interested in reporting LTV:CAC over a period of time in the past, versus optimising marketing spend (by channel) in the future. In the scenario where the goal is reporting you need the average CAC over the reporting period. This would be defined using the standard formula but assuming of course that acquisitions have been carefully estimated.

On the other hand, when it comes to optimisation we are interested in the marginal CAC. Marginal CAC captures the increase in marketing costs by acquiring an additional customer (incrementality). Marketing spend is a convex function of acquisitions and for each additional acquisition, marginal cost grows really fast and so acquiring customers beyond some point can get prohibitively expensive. Marketers are interested in knowing the marginal CAC at various points of demand for customers so they can make comparisons across various channels and decide how to allocate their budget optimally.

Measurement Framework

Before discussing strategies for unbiased measurement, we need to go through the challenges that make measurement hard in the first place. Hopefully, this will provide direction on how to resolve these challenges and while businesses in various industries might have their differences there are still commonalities that will allow us to design a high level marketing measurement playbook.

Challenges in measurement

Tracking limitations

Marketing teams use a variety of online and offline channels to reach out to potential customers. Offline advertising channels like radio and TV have always been hard to track and so Marketing Mix Models (MMMs) have been the go to method for impact measurement. For online channels and since Apple introduced its anti-tracking feature (“Ask app not to track”) tracking has become harder as well.

Correlated channel activity

Marketing teams often deploy advertising campaigns through various channels at the same time to benefit from potential interaction effects. This standard practice introduces what statisticians call multicollinearity in regression analysis that increases the variance of the estimates leading to higher uncertainty over the effectiveness of each channel.

Ad targeting

Channels like paid search introduce selection bias by targeting users that have exhibited high interest in the product and would convert anyway. Let’s assume that your business experiences a big spike in organic acquisitions. A big share of these acquisitions will likely click the more convenient ad that is sitting right above the organic search result. This will overestimate the effect of paid search and underestimate the impact of the organic traffic.

Seasonality

Seasonality can introduce similar biases like the ones we discussed in the previous sections where high correlation appears between seasonal changes in demand and marketing campaigns designed around these seasonal demand spikes. For a dating app that would be advertising on Valentine’s day and for an e-commerce store that would be advertising on Black Friday. If these seasonal changes in demand are not taken into account it could lead the marketer reaching wrong conclusions regarding the source of the acquisitions.

Funnel Effects

Funnel effects can be a symptom of ad targeting and last-click attribution as we have discussed extensively. For example, any impactful offline marketing campaign that precedes paid search could be wrongly attributed to paid search due to last click attribution, making hard to distinguish the impact of these two channels.

The Measurement Playbook

Businesses are not completely powerless against the challenges mentioned above. Let’s go through a comprehensive set of tools that can be used to overcome these challenges.

Experiment design

A lot of the issues mentioned above can be summarised as the correlation between a marketing campaign and other events (other campaigns, holidays, organic changes in demand, etc.). Attempting to break this correlation could make things easier for businesses to measure efficiently the impact of their campaigns. Here are a few examples of how this can be achieved:

• Having full visibility of what is live at any given moment so that their effect can be controlled for during analysis.

• Making sure that new channels can be tested in a way that isolates their effects before attempting to deploy them along with other marketing channels or during seasons of high demand.

• Experiment with increasing intensity in each marketing channel by testing the impact of each channel in acquisitions for various levels of spending, ideally until reaching levels of diminishing returns.

• Repeat experiments either during the same time period to reduce the standard errors of the CAC estimates or at different time periods to ensure that the response curve of each channel is properly calibrated.

Attributable channel strategy

For attributable channels like paid search or social media ads you can try RCTs (e.g. Meta’s conversion lift test, Google Ads conversion lift test) or quasi-experimental methods like geo-tests paired with causal analysis methods like Difference-in-Differences or Synthetic Control and shutdown tests. Paid search platforms allow you to run campaigns on country but also on city level giving you more data to work with. If most of your channels are online, another route would be multi-touch attribution (Markov chain models, game theory).

Let’s say you have an iOS app and you would like to measure the impact that paid brand search has on your acquisitions. By launching the ad at different bidding levels and then shutting it down you can estimate response curves for different levels of spending. In addition, and in the context of the funnel effects we discussed above, we can also estimate the % of acquisitions for varying levels of spending. This allows you to subtract the impact of paid search from total acquisitions and feeding the residual acquisitions to a MMM for further analysis of offline ads. Applying this strategy in the limits of a solid experimentation program can get you a long way.

Non-attributable channel strategy

When it comes to non-attributable channels you still have a few options. The lower hanging fruit would be to deploy an acquisition survey early in the customers journey. This will be useful for a variety of reasons:

• Estimate acquisitions coming from offline channels like OOH, TV, radio.

• Estimate the portion of acquisitions from attributable channels that is coming through other routes and isn’t captured, e.g. ignoring an influencer discount code and becoming a customer in a seemingly organic way. You can then estimate the % of non-attributable acquisitions from an attributable source and assign credit accordingly.

• Build associations between channel specific metrics (e.g. likes) and your response metric (e.g. acquisitions). When the volume of acquisitions from a specific source are high enough to be measurable in the survey, you can draw associations that can be combined with MMMs to give more robust estimates.

The other option would be to build a MMM. By applying rigorous experimentation practices and accounting for funnel effects, seasonality and other control factors, it can be something worth considering. There have been some advances in MMM tech to allow for the utilisation of attribution and experimentation data to calibrate the results of the model and treat selection bias.

References

• Challenges And Opportunities In Media Mix Modeling

• A New Gold Standard for Digital Ad Measurement?

• Calibrating Marketing Mix Models with Experiments and Attribution data

• The Dangerous Seduction of the Lifetime Value (LTV) Formula